Mark Zuckerberg says he'd rather risk 'misspending a couple of hundred billion' than be late to superintelligence (www.businessinsider.com)

Mark Zuckerberg told the "Access" podcast that while an AI bubble is "quite possible," Meta would rather "risk misspending a couple of hundred billion" than be late to a future superintelligence. The company has pledged at least $600 billion for US data centers and infrastructure through 2028 (a figure CFO Susan Li said includes data center buildout and broader US business investments). Zuckerberg framed the strategy as deliberate overcommitment: concentrate elite researchers in a small, flat "superintelligence" lab, remove top-down deadlines, and make "compute per researcher" a competitive advantage by outspending rivals on GPUs and custom infrastructure. The comments matter because they crystallize how big tech is de-risking frontier AI through scale and balance-sheet depth, potentially widening the gap with startups and labs that must continually raise capital (e.g., OpenAI, Anthropic) and are therefore more exposed to market downturns and massive compute bills. Tactically, Meta’s approach emphasizes raw compute, bespoke hardware/software stacks, and talent centralization — levers that accelerate model training and iteration but also concentrate market power and capital risk. For the AI/ML community, that signals continued intense GPU/infra demand, shifting competitive dynamics in where cutting-edge research can be done, and a tradeoff between rapid progress and the economic risks of overbuilding.